Method for Denoising Videos and Electronic Device Therefor

1. A method for denoising videos, applicable to an electronic device comprising a central processing unit (CPU) and a graphics processing unit (GPU), the method comprising: acquiring, by the CPU, video frame images by acquiring video data captured by a user and decoding the video data; loading, by the GPU, the video frame images from the CPU; acquiring, by the GPU, first images by denoising the video frame images using a predetermined non-local means (NLM) denoising algorithm; determining, by the GPU, a noise level of the video frame images; in response to the noise level being greater than or equal to a predetermined noise level threshold, acquiring, by the GPU, second images by denoising the first images using a predetermined non-local Bayes (NLB) denoising algorithm; and acquiring, by the CPU, denoised video data by acquiring the second images from the GPU and encoding the second images, wherein the denoised video data is then shared on a social platform; wherein determining, by the GPU, the noise level of the video frame images comprises: determining, by the GPU, a numerical count of the loaded video frame images; in response to the count of the loaded video frame images being less than a predetermined frame number threshold, determining, by the GPU, noise levels of the video frame images at a predetermined frame interval; and determining, by the GPU, an average of the determined noise levels as the noise level of the video frame images.

2. The method according to claim 1, wherein acquiring, by the GPU, the first images by denoising the video frame images using the predetermined NLM denoising algorithm comprises: allocating, by the GPU, threads to pixels in the video frame images; determining, by the thread, a first reference image patch of the pixel and a corresponding first similar image patch; determining, by the thread, a pixel distance and a coordinate distance of the first similar image patch relative to the first reference image patch; calculating, by the thread, a weight of the first similar image patch based on the pixel distance and the coordinate distance; determining, by the thread, a denoised value of the pixel based on the weight of the first similar image patch; and determining, by the GPU, an image determined based on all denoised pixels as the first image.

3. The method according to claim 2, wherein acquiring, by the GPU, the second images by denoising the first images using the predetermined NLB denoising algorithm comprises: allocating, by the GPU, threads to pixels in the first images; determining, by the thread, a second reference image patch of the pixel in the first image and a corresponding second similar image patch; calculating, by the thread, a covariance matrix of the second reference image patch based on the second similar image patch and the second reference image patch; calculating, by the thread, a mean value of the first reference image patch based on the first reference image patch and the first similar image patch; calculating, by the thread, denoised pixel values of the second reference image patch based on the covariance matrix of the second reference image patch and the mean value of the first reference image patch; determining, by the thread, a denoised value of the pixel in the first image based on all denoised second reference image patches comprising the pixel; and determining, by the GPU, an image determined based on all denoised pixels as the second image.

4. The method according to claim 3, wherein upon allocating, by the GPU, the threads to the pixels in the first images, the method further comprises: segmenting, by the GPU, the first image into a plurality of second sub-images; configuring, by the GPU, the threads corresponding to the pixels belonging to a same second sub-image to a same thread block; allocating, by the GPU, data required for executing the NLB denoising algorithm to the thread block; and in the case that the threads in the same thread block are executed in terms of warps, cooperatively reading, by the threads in the warps, the required data into a shared memory of the GPU.

5. The method according to claim 3, wherein determining, by the thread, the second reference image patch of the pixel in the first image and the corresponding second similar image patch comprises: determining, by the thread, the second reference image patch by taking the pixel as a center; acquiring, by the thread, a corresponding second search box from a shared memory of the GPU; and determining, by the thread, the second similar image patch corresponding to the second reference image patch in the second search box at a predetermined second pixel interval.

6. The method according to claim 2, wherein upon allocating, by the GPU, the threads to the pixels in the video frame images, the method further comprises: segmenting, by the GPU, the video frame image into a plurality of first sub-images; configuring, by the GPU, the threads corresponding to the pixels belonging to a same first sub-image to a same thread block; allocating, by the GPU, data required for executing the NLM denoising algorithm to the thread block; and in the case that the threads in the same thread block are executed in terms of warps, cooperatively reading, by the threads in the warps, the required data into a shared memory of the GPU.

7. The method according to claim 2, wherein determining, by the thread, the first reference image patch of the pixel and the corresponding first similar image patch comprises: determining, by the thread, the first reference image patch by taking the pixel as a center; acquiring, by the thread, a corresponding first search box from a shared memory of the GPU; and determining, by the thread, the first similar image patch corresponding to the first reference image patch in the first search box at a predetermined first pixel interval.

8. The method according to claim 1, wherein determining, by the GPU, the noise level of the video frame images further comprises: in the case that the number of the loaded video frame images is greater than or equal to the predetermined frame number threshold, selecting, by the GPU, a predetermined number of video frame images from the loaded video frame images; determining, by the GPU, noise levels of the video frame images in the predetermined number of video frame images at the predetermined frame interval; and determining, by the GPU, an average of the determined noise levels as the noise level of the video frame images.

9. The method according to claim 1, wherein upon acquiring, by the CPU, the video frame images by decoding the video data, the method further comprises: filling, by the CPU, the video frame images.

10. A non-transitory computer-readable storage medium storing a computer program, wherein the computer program, when loaded and run by a processor of a device, causes the device to perform processes of the method for denoising videos as defined in claim 1.

11. A method for denoising videos, applicable to an electronic device comprising a central processing unit (CPU) and a graphics processing unit (GPU), the method comprising: acquiring, by the CPU, video frame images by acquiring video data captured by a user and decoding the video data; loading, by the GPU, the video frame images from the CPU; determining, by the GPU, a noise level of the video frame images; acquiring, by the GPU based on the noise level, first images by denoising the video frame images using a predetermined first image denoising algorithm, or second images by acquiring the first images by denoising the video frame images using the predetermined first image denoising algorithm and denoising the first images using a predetermined second image denoising algorithm; and acquiring, by the CPU, denoised video data by acquiring the first images from the GPU and encoding the first images, or by acquiring the second images from the GPU and encoding the second images, wherein the denoised video data is then shared on a social platform; wherein determining, by the GPU, the noise level of the video frame images comprises: determining, by the GPU, a numerical count of the loaded video frame images; in response to the count of the loaded video frame images being less than a predetermined frame number threshold, determining, by the GPU, noise levels of the video frame images at a predetermined frame interval; and determining, by the GPU, an average of the determined noise levels as the noise level of the video frame images.

12. The method according to claim 11, wherein acquiring, by the GPU based on the noise level, the first images by denoising the video frame images using the predetermined first image denoising algorithm, or the second images by acquiring the first images by denoising the video frame images using the predetermined first image denoising algorithm and denoising the first images using the predetermined second image denoising algorithm comprises: in the case that the noise level is less than a predetermined noise level threshold, acquiring, by the GPU, the first images by denoising the video frame images using the predetermined first image denoising algorithm; and in the case that the noise level is greater than or equal to the predetermined noise level threshold, acquiring, by the GPU, the second images by acquiring the first images by denoising the video frame images using the predetermined first image denoising algorithm and denoising the first images using the predetermined second image denoising algorithm.

13. The method according to claim 12, wherein acquiring, by the GPU, the first images by denoising the video frame images using the predetermined first image denoising algorithm comprises: allocating, by the GPU, threads to pixels in the video frame images; determining, by the thread, a first reference image patch of the pixel and a corresponding first similar image patch; determining, by the thread, a pixel distance and a coordinate distance of the first similar image patch relative to the first reference image patch; calculating, by the thread, a weight of the first similar image patch based on the pixel distance and the coordinate distance; determining, by the thread, a denoised value of the pixel based on the weight of the first similar image patch; and determining, by the GPU, an image determined based on all denoised pixels as the first image.

14. The method according to claim 13, wherein acquiring, by the GPU, the second images by denoising the first images by using the predetermined second image denoising algorithm comprises: allocating, by the GPU, threads to pixels in the first images; determining, by the thread, a second reference image patch of the pixel in the first image and a corresponding second similar image patch; calculating, by the thread, a covariance matrix of the second reference image patch based on the second similar image patch and the second reference image patch; calculating, by the thread, a mean value of the first reference image patch based on the first reference image patch and the first similar image patch; calculating, by the thread, denoised pixel values of the second reference image patch based on the covariance matrix of the second reference image patch and the mean value of the first reference image patch; determining, by the thread, a denoised value of the pixel in the first image based on all denoised second reference image patches comprising the pixel; and determining, by the GPU, an image determined based on all denoised pixels as the second image.

15. An electronic device for denoising videos, comprising a central processing unit (CPU), a graphics processing unit (GPU), a memory, and a computer program stored in the memory and runnable on the CPU and the GPU, wherein the CPU and the GPU, when loading and running the computer program, is caused to perform processes of a method for denoising videos as defined in claim 11.

16. A non-transitory computer-readable storage medium storing a computer program, wherein the computer program, when loaded and run by a processor of a device, causes the device to perform processes of the method for denoising videos as defined in claim 11.

17. An electronic device for denoising videos, comprising a central processing unit (CPU), a graphics processing unit (GPU), a memory, and a computer program stored in the memory and runnable on the CPU and the GPU, wherein the CPU and the GPU, when loading and running the computer program, are caused to perform processes of a method for denoising videos, wherein the method comprises: acquiring, by the CPU, video frame images by acquiring video data captured by a user and decoding the video data; loading, by the GPU, the video frame images from the CPU; acquiring, by the GPU, first images by denoising the video frame images using a predetermined non-local means (NLM) denoising algorithm; determining, by the GPU, a noise level of the video frame images; in response to the noise level being greater than or equal to a predetermined noise level threshold, acquiring, by the GPU, second images by denoising the first images using a predetermined non-local Bayes (NLB) denoising algorithm; and acquiring, by the CPU, denoised video data by acquiring the second images from the GPU and encoding the second images, wherein the denoised video data is then shared on a social platform; wherein determining, by the GPU, the noise level of the video frame images comprises: determining, by the GPU, a numerical count of the loaded video frame images; in response to the count of the loaded video frame images being less than a predetermined frame number threshold, determining, by the GPU, noise levels of the video frame images at a predetermined frame interval; and determining, by the GPU, an average of the determined noise levels as the noise level of the video frame images.

18. The electronic device according to claim 17, wherein acquiring, by the GPU, the first images by denoising the video frame images using the predetermined NLM denoising algorithm comprises: allocating, by the GPU, threads to pixels in the video frame images; determining, by the thread, a first reference image patch of the pixel and a corresponding first similar image patch; determining, by the thread, a pixel distance and a coordinate distance of the first similar image patch relative to the first reference image patch; calculating, by the thread, a weight of the first similar image patch based on the pixel distance and the coordinate distance; determining, by the thread, a denoised value of the pixel based on the weight of the first similar image patch; and determining, by the GPU, an image determined based on all denoised pixels as the first image.